built a weird little library last friday night. it uses LLMs to auto-generate JavaScript class methods at runtime; no need to define the method ahead of time—just decorate your class, call whatever you want, and the model figures it out on the fly.
it uses `Proxy`s, and pipes the class definition and the call example to the LLM for it to generate a valid method.
of course it was not built for production use, but it was a fun exploration. would love to hear your thoughts!
It'd be helpful if they exposed a summary of the chain-of-thought response instead. That way they'd not be leaking the actual tokens, but you'd still be able to understand the outline of the process. And, hopefully, understand where it went wrong.
It's more useful to fix the problem before it happens. Try to set up systems or workflows that make it easy to do the right thing, rather than trying to fix it afterwards.
I really had to converse with the lead in this wise:
If I stapled your HR record, including your bank account and routing number, to our shared refrigerator so I could remember it, would you think that appropriate?
Really, this was the context of our meeting.
Now I am left with the aftermath of his departure, and knives at my sides.
Just to add to the reply, the AdS/CFT correspondence (aka Maldacena duality) was proposed by Juan Maldacena, one of the authors of this paper (Humanly traversable wormholes).
I felt the same way about Netflix when it first started. Now, there are dozens of competitors with different catalogues, making piracy enticing for people once again.
Speaking as a consumer, that in itself doesn't make it enticing. If there was very little worthwhile content on 1-2 platforms, maybe, but since I'm not the "binging" type who absolutely needs to see every hyped show right now, I can't be arsed and find there's already enough to see.
Between Netflix and Prime, you're not spending that much and you sure as hell don't need a dozen subscriptions to watch every good show.
I think there are still enough people who remember 2010-era Starz on Netflix (and with the massive DVD catalogue for the rest). Everything, on every single service, has been downhill since then.
I'm the same as you. I subscribe to Netflix so long as there's something I'm watching, then stop. Then I might subscribe to something else. People that complain about having to pay for all services are just power users, and I think it's fair that I don't pay as much as them.
I recommend you the story "The Last Question", by Isaac Asimov[0], one of my favourite stories of all time. It's about how the question about reverting the direction of the entropy keeps rising on people, throughout the life of the universe.
My country (Argentina) has a lottery(esque?) approach to giving work visas for the US, so basically: even with company support, work visas are subject to probability (~30% chance of getting it for a given year). Do you have experience, or do you know, how companies handle hires from countries with this methodology?
The next H-1B lottery will take place in March 2021. The proclamation suspended the issuance of new H-1B visas; the lottery hasn't been affected and those in the U.S. seeking to change their status to H-1B or extend their H-1B status haven't been affected.
it uses `Proxy`s, and pipes the class definition and the call example to the LLM for it to generate a valid method.
of course it was not built for production use, but it was a fun exploration. would love to hear your thoughts!